Generative Fill with AI and 3D
Hey all, You've probably seen projects that add objects to an image from a style or text prompt, like InteriorAI (levelsio) and Adobe Firefly. The prevalent issue with these diffusion-based inpainting approaches is that they don't yet have great conditioning on lighting, perspective, and structure. You'll often get incorrect or generic shadows; warped-looking objects; and distorted backgrounds. What is Fill 3D? Fill 3D is an exploration on doing generative fill in 3D to render ultra-realistic results that harmonize with the background image, using industry-standard path tracing, akin to…
In plain words
Fill 3D is a generative image tool that adds objects to photos by working in three dimensions rather than two. Unlike existing text-based inpainting tools, it uses 3D rendering and path tracing to ensure added objects match the original image's lighting, perspective, and shadows realistically. Users draw rectangles on their image, describe what they want added, and the tool generates photorealistic results that blend seamlessly with the background, similar to professional compositing techniques used in film production.
written from the facts on this page · September 2026
From the sources
In the maker’s words, at launch
Hey all, You've probably seen projects that add objects to an image from a style or text prompt, like InteriorAI (levelsio) and Adobe Firefly. The prevalent issue with these diffusion-based inpainting approaches is that they don't yet have great conditioning on lighting, perspective, and structure. You'll often get incorrect or generic shadows; warped-looking objects; and distorted backgrounds. What is Fill 3D? Fill 3D is an exploration on doing generative fill in 3D to render ultra-realistic results that harmonize with the background image, using industry-standard path tracing, akin to compositing in Hollywood movies. How does it work? 1. Deproject: First, deproject an image to a 3D shell using both geometric and photometric cues from the input image. 2. Place: Draw rectangles and describe what you want in them, akin to Photoshop's Generative Fill feature. 3. Render: Use good ol' path tracing to render ultra-realistic results. Why Fill 3D? + The results are insanely realistic (see video in the github repo, or on the website). + Fast enough: Currently, generations take 40-80 seconds. Diffusion takes ~10seconds, so we're slower, but for the level of realism, it's pretty good. + Potential applications: I'm thinking of virtual staging in real estate media, what do you think? Check it out at https://fill3d.ai + There's API access! :D + Right now, you need an image of an empty room. Will loosen this restriction over time. Fill 3D is built on Function (https://fxn.ai). With Function, I can run the Python functions that do the steps above on powerful GPUs with only code (no Dockerfile, YAML, k8s, etc), and invoke them from just about anywhere. I'm the founder of fxn. Tell me what you think!! PS: This is my first Show HN, so please be nice :)
More ai this month
the category →
I trained a 125M-parameter transformer to autocomplete piano performances in real time (~108 notes/sec on an iPhone 15). The idea is basically GitHub Copilot or Tabnine, except instead of prompting it with code, you prompt it by playing a few notes on a MIDI piano. The model then continues what you played, entirely on-device. The app is free if anyone wants to try it. Happy to answer questions about the model, training, Core ML, or the many things that didn't work.
AI · 17d ago · simedw.com
Astute▲585Automate your B2B brand going viral, with new media creators
AI · 18d ago · company-app.joinastute.com
Hey HN, Henry from Cactus here! We previously released Cactus Needle, a 14MB agentic LLM for tool call, device use, and structured extraction for phones, wearables, smart homes, small robots and microcontrollers. We got really great feedback here, and have now incorporated the suggestions to release Needle 2. The whole model is a single 14MB binary that runs a full session in 28MB of RAM; 45m parameters at 2bit compression. Needle hits 500 tokens/sec decode speed on a Raspberry Pi 5, sits between 400-1,500 tokens/sec on VR devices like Meta Quest 3S and Apple Vision Pro, and ranges…
AI · 27d ago · cactuscompute.com
